Authors Niamjit Das of Shahjalal University of Science and Technology and the Center for Natural Resource Studies in Bangladesh, together with Kevin Andrew of Stanford University’s Department of Earth System Science, conducted the review under PRISMA 2020 guidelines. They synthesized 57 peer-reviewed studies published between 2016 and 2025 that combined Earth Observation datasets with ecosystem modeling and assimilation frameworks. Across these studies, 480 instances of remote sensing platforms were reported, with a mean of 8.4 per study, alongside 340 ecosystem modeling approaches and 132 forest dynamics contexts. The review was guided by a central question: how can Digital Twin Forestry integrate multisensor data, ecosystem models, and data assimilation to move toward near-real-time, uncertainty-aware forecasting for climate-resilient forest management?

The authors define Digital Twin Forestry as more than conventional model-data integration. A true digital twin must satisfy several minimum characteristics: representation of a physical forest system through a virtual counterpart, repeated or continuous observational updates from multisensor platforms, dynamic updating of model states or parameters in response to new data, explicit propagation and calibration of uncertainty, and feedback loops that connect forecasts to decision-support outputs. Systems that involve only static monitoring or episodic data assimilation are enabling technologies, not full digital twins. This operational definition builds on broader digital twin concepts articulated across engineering and Earth-system domains but specifies minimum criteria for forestry applications, including continuous observational updates, dynamic model coupling, uncertainty propagation, and feedback to decision support.

The observational backbone of the framework is a rapidly evolving suite of remote sensing platforms. Small satellite constellations such as PlanetScope and SkySat provide daily 3 to 5 meter imagery for disturbance and recovery mapping, though limited spectral depth constrains trait retrieval. Hyperspectral missions such as PRISMA and EnMAP, operating at roughly 30 meter resolution with more than 100 bands, derive foliage biochemical constituents including chlorophyll and lignin, enabling robust estimates of canopy nitrogen and pigment content. Planned missions such as NASA SBG and ESA CHIME, along with solar-induced fluorescence products from OCO-2 and OCO-3, are expected to enhance assimilation of photosynthetic processes and reduce Gross Primary Productivity error during disturbance events.

Active structural sensors are equally indispensable. NASA GEDI captures canopy height and profiles at approximately 25 meter resolution, reducing biomass retrieval errors compared to optical indices. ICESat-2 measures forests via photon-counting laser paths, while airborne and UAV LiDAR provide flexible resolutions for biodiversity, canopy coverage, and biomass estimation, including REDD+ applications. Spaceborne SAR missions such as NISAR and the SAR Constellation monitor structural shifts and soil-moisture variations under cloudy conditions. Joint SAR-LiDAR use significantly reduces RMSE in tropical forest biomass estimation, demonstrating the benefits of multi-sensor fusion. At finer scales, UAV-mounted multispectral and hyperspectral platforms deliver centimeter-scale trait and stress detection, supporting early forest-health monitoring and site-specific management, while proximal IoT arrays with radiometers, thermal cameras, and micrometeorological stations provide near-real-time ground-layer measurements.

Ecosystem models form the computational core of the framework. Land-surface models such as CLM and JULES simulate photosynthesis, respiration, transpiration, and carbohydrate allocation at daily to sub-daily timescales, and the inclusion of plant hydraulic schemes has improved forecasts of drought thresholds and productivity loss. Distributed dynamic global vegetation models such as LPJ-GUESS and ORCHIDEE extend these processes to Earth-system contexts, supporting biome-scale projections and national carbon inventories. Trait-based models such as JeDi and aDGVM2 represent species within communities by linking functional traits to carbon storage and drought resistance, providing mechanistic insights into resilience. Landscape models such as LANDIS-II simulate disturbance, succession, and management across heterogeneous terrain, facilitating fire dynamics and connectivity analyses, while hybrid approaches combine mechanistic modules with machine learning emulators and ensemble pipelines integrate diverse models to reveal consistent vulnerability patterns.

Data assimilation provides the essential informatics bridge between multisensor observations and ecosystem model outputs, producing physically consistent, observation-constrained state estimates and forecasts. Ensemble Kalman Filters propagate forecasts and error covariance through ensembles, updating states with observations in near-real-time; demonstrated applications include assimilation of MODIS LAI and FAPAR into CLM, which reduced phenology bias and improved carbon-flux estimates compared to open-loop simulations. Particle Filters approximate non-Gaussian posteriors using weighted ensembles, effectively capturing threshold processes such as die-off and regeneration; assimilation of OCO-2 solar-induced fluorescence during the 2015-2016 El Niño improved GPP estimates and captured drought-induced declines in photosynthesis that ensemble methods missed. Four-dimensional variational methods minimize cost functions over time windows using tangent-linear and adjoint operators, and hybrid ensemble-variational systems blend flow-dependent ensemble covariance with static variational covariance, improving forecasts of canopy dynamics under disturbances such as fire or pests.

The review documents compelling case studies that illustrate the framework’s potential. OCO-2 solar-induced fluorescence captured diverse impacts across Amazon ecoregions during the 2015-2016 El Niño, revealing high spatial variability, and these measurements have been assimilated into process-based models via ensemble or particle filters, dynamically updating GPP predictions in near-real-time. Post-fire recovery in the western United States has been evaluated using fused Sentinel-2 and PlanetScope reflectance data with GEDI LiDAR canopy metrics, assimilated into succession models to update recovery trajectories. UAV-based hyperspectral systems detect early pest outbreaks by recording canopy-stress signatures at centimeter resolution, and within prototype frameworks these stress signals have been assimilated into trait-based demographic models or hybrid machine learning surrogates, updating predictions of mortality risk and canopy decline. Fusion of satellite LAI and biomass products has enhanced global carbon accounting, and airborne LiDAR and thermal sensors have been used to map urban forests, quantifying canopy structure, biomass, and cooling services.

Yet significant challenges remain. The 57 included studies were unevenly distributed across biomes: 21 percent in temperate forests, 13 percent in tropical forests, 10 percent in boreal forests, 9 percent in mangroves, 5 percent in urban forests, 28 percent spanning global or multi-biome contexts, and 14 percent general forest studies. Tropical and boreal regions remain underrepresented due to persistent cloud cover, limited local sensor deployment, and logistical constraints, highlighting broader equity concerns, as regions with the greatest biodiversity and climate vulnerability often face the greatest monitoring challenges. A chi-square test confirmed a non-random distribution of platforms across biomes, with SAR data used disproportionately in tropical biomes where cloud cover limits optical sensing, while optical platforms dominated temperate forest studies. Country affiliation networks showed that most outputs originated from North America, Europe, and China, with tropical nations notably underrepresented, reflecting disparities in sensor availability, computational infrastructure, and research investment.

Informatics bottlenecks also constrain the framework from becoming fully operational. Earth Observation products remain affected by retrieval errors, atmospheric interference, sensor drift, and platform-specific biases that propagate through model initialization and assimilation. Scale mismatches between high-resolution observations at meter to sub-meter scales and coarser model grids of at least 1 kilometer obscure canopy structure and species composition at fine scales. Many models remain constrained by simplified plant functional type schemes and legacy parameterization, limiting predictive accuracy under novel climate regimes. Assimilating large, high-frequency observation datasets is resource-intensive, with satellite constellations and UAV campaigns generating petabyte-scale archives, while advanced assimilation techniques require repeated forward-model runs. Interoperability remains a major bottleneck, with incompatible file types, metadata standards, and limited APIs hindering reproducibility and innovation, and high-resolution data, model outputs, and high-performance computing remain concentrated in well-resourced institutions while biodiversity hotspots in the Global South face persistent infrastructure and knowledge gaps.

Future priorities identified by the authors include advancing physics-informed machine learning, decentralized federated learning, deeper integration of socio-ecological governance drivers, and standardized protocols to ensure interoperability across platforms and scales. Embedding biodiversity and resilience monitoring into near-real-time, uncertainty-aware workflows will be critical, enabling Digital Twin Forestry to evolve from prototype demonstrations toward operational systems that provide adaptive, policy-relevant decision support. The framework contributes to global sustainability agendas, supporting the United Nations Sustainable Development Goals including Climate Action and Life on Land. Most forestry digital twins remain at prototype or early operational stages, requiring further validation, benchmarking, and scaling before widespread deployment, but the review offers an integrative roadmap for advancing the paradigm, contributing to global sustainability and supporting climate-resilient forest management aligned with international biodiversity and climate frameworks.

Subject of Research: Environmental Informatics

Article Title: Digital twin forestry as an informatics framework integrating remote sensing ecosystem modelling and data assimilation

Article References: Das, N., & Andrew, K. (2026). Digital twin forestry as an informatics framework integrating remote sensing ecosystem modelling and data assimilation. Discover Informatics, 1(1), Article 29. https://doi.org/10.1007/s44564-026-00029-8

Image Credits: AI Generated

DOI: 10.1007/s44564-026-00029-8

Keywords: Digital Twin Forestry, Remote Sensing, Ecosystem Modeling, Data Assimilation, Climate-Resilient Forest Management, Earth Observation, Machine Learning, Biodiversity Monitoring, FAIR Principles, Sustainable Development Goals, Digital, twin